Long-Term EEG Monitoring

Routine EEG is a thirty-minute snapshot; long-term monitoring (LTM) is a film. By recording continuously for hours to days with time-locked video, the epilepsy monitoring unit (EMU) answers questions a snapshot cannot: are these spells epileptic at all, how do they begin clinically and electrographically, and - for the surgical candidate - where does the seizure start? The defining feature is the synchronization of behavior and electrography. A reviewer can scroll to the second a patient's hand begins to fidget and read the EEG at that instant, then watch the clinical sequence unfold against the evolving rhythm. This pairing is what transforms LTM from a longer EEG into a fundamentally different diagnostic instrument, and it is the reason the EMU remains the reference standard against which every ambulatory and wearable seizure-detection technology is measured.

Video-EEG and the capture of habitual events

The first and often most valuable yield of video-EEG is classification of paroxysmal events. A substantial fraction of patients referred to an EMU for apparently refractory epilepsy are found instead to have psychogenic nonepileptic seizures (PNES) or, less often, syncope, parasomnias, or movement disorders. The diagnosis of PNES is made positively, not by exclusion. The ILAE staged scheme grades diagnostic certainty from possible, through probable and clinically established, to documented - and a typical event captured on video-EEG with characteristic semiology and no ictal or postictal correlate represents that highest, documented tier. Communicating the diagnosis is itself a clinical skill: framing PNES as a real and treatable condition, rather than as the absence of epilepsy, materially improves engagement with the psychological treatment that actually helps.

The semiologic features that favor PNES carry very different evidential weight, and a sophisticated reader treats them as likelihood ratios rather than as a checklist. The strongest single sign is ictal eye closure, often with forced closure and active resistance to opening - epileptic seizures characteristically begin with the eyes open. Other features that shift probability toward PNES include asynchronous, out-of-phase limb movements, ictal weeping or crying, a fluctuating, waxing-and-waning course, and unusually long duration. A second group - side-to-side head shaking, postictal crying, and preserved awareness or recall during bilateral motor activity - is highly specific but insensitive: useful when present, uninformative when absent. Two cautions complete the picture. Pelvic thrusting has limited discriminatory value because it also occurs in frontal-lobe and convulsive seizures, and the classic 'arc de cercle' back-arching does not reliably discriminate in EMU data. And no single sign is pathognomonic; the diagnosis rests on the constellation together with the EEG.

Beyond the binary question of epileptic versus non-epileptic, the EMU refines classification in ways that change treatment. Capturing habitual events can reveal that a patient labeled with focal epilepsy in fact has a genetic generalized syndrome - or the reverse - a distinction that determines which antiseizure medications will help and which may worsen seizures, since several agents effective in focal epilepsy aggravate generalized epilepsies. Prolonged recording can also expose a second, independent seizure type that the history missed, quantify a true seizure frequency in a patient whose reporting is unreliable, and disentangle the overlap in which epilepsy and psychogenic events coexist in the same patient, a combination common enough that finding a non-epileptic event never ends the search for an epileptic one. None of this is possible without the safety architecture that makes prolonged provoked recording acceptable: a dedicated unit with continuous observation, trained staff who can intervene during a convulsion, seizure-detection alarms, padded surroundings, and explicit rescue and status-epilepticus protocols. The diagnostic power of the EMU and its physical safety design are inseparable, because the events it exists to capture are, by their nature, the events that can harm the patient.

When events are epileptic, video-EEG documents the semiology - the lawful clinical sequence of a seizure - which itself carries localizing information independent of the EEG. The unit captures the ictal onset, the propagation, and the postictal state, building a library of the patient's habitual seizures. Frontal-lobe seizures are the great mimic and the great trap here: they can be brief, hyperkinetic, bizarre, nocturnal, and accompanied by preserved awareness, with a scalp-invisible onset, so they imitate PNES on both video and EEG and demand particular caution before either label is applied. Spot review is supplemented by automated seizure and spike detection software and, increasingly, by machine-learning event classifiers, but these algorithms are sensitive aids, not arbiters; the human reviewer confirms every flagged event and, just as importantly, reviews patient- and nurse-triggered event buttons against the record, because the alignment of a pushbutton with the electrographic onset is itself diagnostic information.

Seizure-capture strategies and withdrawal risk

The central tension of the EMU is that its purpose - recording seizures - requires provoking them safely. The most powerful and most dangerous tool is antiseizure medication (ASM) withdrawal. Tapering or stopping a patient's medications increases seizure frequency and shortens the time to first seizure, but it also raises the probability of secondarily generalized convulsions, of seizure clustering, and of status epilepticus, and it can distort the very data being collected: withdrawal can recruit seizures from regions that are not the habitual focus, producing falsely multifocal, contralateral, or non-localized onsets. The art is to taper only as fast as necessary, with continuous nursing observation, a written rescue protocol, intravenous access at the ready, and explicit attention to the risk of a generalized convulsion in a patient whose medications have been removed - a setting in which postictal events and even SUDEP-relevant physiology are concentrated.

Capture strategy is dictated by purpose. For diagnostic monitoring - the question being whether the events are epileptic at all - minimal provocation and patience are preferred, because the goal is simply to record a habitual event in its natural form. For presurgical monitoring the goal is different: to record several of the patient's typical seizures so that onset can be defined reliably, which justifies more assertive provocation - but only enough to capture the habitual event, never to chase a target number. Recording too few seizures risks a spurious localization built on a single atypical event; recording under excessive withdrawal risks the opposite error of a contaminated, non-habitual onset. The number of seizures required is a judgment about consistency: a tightly stereotyped patient with concordant data may be defined by two or three habitual seizures, whereas an inconsistent or discordant picture demands more.

Presurgical evaluation and the cortical zones

Epilepsy surgery rests on a single conceptual target: the epileptogenic zone, the region of cortex necessary and sufficient for seizure generation - the cortex that must be removed or disconnected to render the patient seizure-free. The epileptogenic zone is a theoretical construct that cannot be measured directly and, strictly, can only be defined retrospectively: if the patient becomes seizure-free after resection, the removed tissue contained it. This logical structure matters, because a surgical failure tells you the resection was insufficient but not where the true zone lay. Because the epileptogenic zone cannot be observed prospectively, it is estimated by triangulating several imperfect surrogate zones, the framework articulated by Rosenow and Luders. The ictal-onset (seizure-onset) zone is the cortex from which the recorded seizures begin electrographically. The irritative zone generates interictal spikes and is typically larger than, and contains, the onset zone. The symptomatogenic zone produces the first ictal symptoms read from semiology - and it may be remote from onset, because a seizure can begin in clinically silent cortex and only declare itself when it propagates into eloquent areas.

Two further zones complete the map. The epileptogenic lesion is the causal structural abnormality on MRI or pathology. The functional deficit zone is the region of interictal, non-epileptic dysfunction, revealed by the neurological examination, by neuropsychological testing, by interictal slowing on EEG, and by interictal hypometabolism on FDG-PET. The original framework adds a sixth region that is conceptually different in kind - the eloquent cortex, the tissue subserving language, primary motor, primary visual, and memory function that must be preserved rather than localized for resection. The surgical plan is the intersection of two maps: where the seizures come from, and what must not be touched.

The relationship between the irritative zone and the seizure-onset zone deserves particular attention because it is where scalp EEG most often misleads. Interictal spikes mark the irritative zone, which is typically broader than the onset zone and frequently extends beyond it - so a resection guided by interictal spikes alone risks either removing too much eloquent tissue or, paradoxically, missing the true onset when the dominant spike population sits downstream of where seizures actually begin. This is why ictal recording is indispensable and why the electrographic onset pattern itself carries information. Focal seizures often begin not with a slow rhythmic discharge but with low-voltage fast activity or a brief electrodecremental attenuation at onset, patterns that are highly localizing when captured intracranially but are frequently invisible on the scalp, where the seizure may only become apparent after it has propagated and recruited a larger field. The lag between true onset and scalp expression is one of the central reasons a discordant or non-localized scalp ictal onset triggers intracranial evaluation: the scalp may simply be showing propagation rather than origin, and only electrodes placed at the candidate source can resolve the difference between where a seizure is first seen and where it is actually born.

The decisive operating principle is concordance. A favorable surgical candidate is one in whom these independent lines of evidence converge on the same region: a focal MRI lesion, interictal spikes over it, a stereotyped semiology pointing to it, and an ictal onset arising from it. The canonical example is mesial temporal lobe epilepsy with hippocampal sclerosis on MRI, concordant anterior temporal interictal and ictal EEG, and a semiology of focal impaired-awareness seizures with automatisms - a constellation that predicts a high probability of seizure freedom after anterior temporal lobectomy. The historical benchmark is the Wiebe randomized trial, in which 58 percent of surgical patients versus 8 percent of medically treated patients were free of awareness-impairing seizures at one year. A conservative statement of the durable outcome is that Engel class I (seizure freedom) is achieved in roughly 60-70 percent of well-selected temporal cases at one to two years, with a real decline over the following decade to approximately 50-60 percent - a candor about late relapse that honest preoperative counseling requires.

Discordance is the warning sign. A normal MRI (so-called nonlesional epilepsy), bilateral or non-localized scalp ictal onsets, or a semiology that contradicts the EEG localization all lower confidence and typically trigger the next tier of evaluation: intracranial recording. Here the field has shifted. Stereo-electroencephalography (sEEG) - depth electrodes placed stereotactically through small drill holes - has increasingly displaced subdural grids, particularly in North America, because it samples deep, mesial, insular, and bilateral structures that grids cannot reach and avoids a large craniotomy. The evidence supports a specific safety advantage rather than a blanket one: sEEG has a clearly lower infection rate (on the order of 0.3 versus 1.8 percent), with hemorrhage rates comparable to grids and an overall complication trend that favors sEEG without reaching significance in pooled data. Subdural grids retain an advantage for contiguous superficial cortical coverage and for extraoperative functional mapping. Diagnostic yield and seizure-freedom outcomes are broadly similar between the two; there is no randomized comparison, the literature is observational and subject to confounding by indication, and institutional expertise probably matters more than the modality itself.

High-frequency oscillations and the discipline of unvalidated biomarkers

The search for a direct electrographic marker of epileptogenic tissue has centered on high-frequency oscillations (HFOs) - brief bursts in the ripple (about 80-250 Hz) and fast-ripple (about 250-500 Hz) bands recorded intracranially. The appeal is conceptual: HFOs, and fast ripples especially, are more spatially specific to the seizure-onset zone than interictal spikes, raising the hope that resecting HFO-generating tissue might tailor surgery better than existing markers. For a PhD-level reader, however, the more important lesson is epistemic, and it is a cautionary one. As of 2026, HFOs remain investigational and are not validated for routine clinical decision-making, and the reason is instructive about how candidate biomarkers should be judged.

The decisive datum is that the only randomized trial of HFO-guided surgery - a single-center, single-blind, adaptive non-inferiority trial reported in 2022 - failed to demonstrate non-inferiority of HFO-guided resection compared with spike-guided resection; seizure-freedom was lower in the HFO-guided arm (roughly 67 versus 90 percent), with the disadvantage concentrated in temporal-lobe cases and the extratemporal results inconclusive. A prospective multicenter study had earlier found HFO-based prediction correct in only about 69 percent of patients, with substantial variability between centers and several patients rendered seizure-free despite most HFO-bearing tissue being left in place. The mechanistic obstacles are real: physiological HFOs occur in normal brain and overlap with pathological HFOs so that frequency alone cannot separate them; there is no universally accepted amplitude or rate threshold; and automated detection is poorly reproducible across reviewers and centers. Retrospective meta-analyses do show an association between complete resection of HFO-generating tissue and seizure freedom, but retrospective association is not prospective validation - and the one randomized test was negative. The honest framing is that this is a textbook example of a biologically plausible biomarker that has not survived prospective evaluation, and that the prevailing view rests on the literature and that single trial rather than on any formal society guideline declaring HFOs ready or unready.

Finally, localization is never separated from function or from risk. Resection must spare eloquent cortex, so the presurgical workup pairs localization of the epileptogenic zone with mapping of language, motor, visual, and memory function. Functional MRI has largely replaced the intracarotid amobarbital (Wada) procedure for language lateralization in many centers, though it is less robust for memory, so Wada persists for memory assessment and for discordant or high-risk cases - and recurrent amobarbital shortages have driven the use of alternative agents such as methohexital, etomidate, and propofol. Direct electrical cortical stimulation remains the gold standard for mapping eloquent cortex. Surgery is also offered against the backdrop of SUDEP, whose risk is dominated by frequent generalized tonic-clonic seizures and is concentrated in exactly the refractory population referred for surgical evaluation - a risk gradient that makes the prospect of seizure freedom not merely a quality-of-life goal but a potentially life-prolonging one. The value of long-term monitoring is precisely that it produces, with a rigor no shorter study can match, the first of the two maps - where the seizures come from - on which every one of these decisions depends.